Table 4.
Summary of error profiles across participants.
| Distortions |
Categorical |
||
|---|---|---|---|
| Parameter | Value | Parameter | Value |
| Total number of errors | 372 | Total number of errors | 143 |
| One phoneme-matched stimulus a | 341 | Perceived as one of six canonical syllables a | 81 |
| Classified as b | Classified as b | ||
| Stimulus only | 47% | Stimulus only | 5% |
| One other only | 16% | Perceived only | 43% |
| Stimulus plus one other | 18% | Stimulus and perceived only | 20% |
| Multiple | 1% | Other combination or multiple | 18% |
| None | 18% | None | 15% |
Note. Shown are the total counts of errors of various types (across participants) and the percentage of errors that were classified into different error profiles using a quadratic discriminant analysis classifier.
Productions that contained a combination of distortions and categorical errors (e.g., target /du/, transcribed /b@/) or in which the entire syllable was distorted (/@/) and categorical errors that were transcribed as a legal syllable but not one of the six in the entire stimuli set (e.g., /di/) were removed, as there were no canonical, nonerror productions to compare against.
Errors were classified via six classifiers per participant, each trained to recognize one syllable as in class and all other syllables as out of class. These were then used to create error profiles. For example, if a distortion error were classified as in class for only the stimulus syllable and out of class for the remaining syllables, it was denoted here in the left column as stimulus only.